Artificial Neural Network and Fuzzy Logic Evaluation for Helling Up and Non-helling Up Furrows Infiltration Simulation in Sugarcane Farms

Sugarcane (Saccharum.spontaneum L.) is one of the high consumption water plants which is irrigated by both helling up (H) and non-helling up (N) furrow irrigation methods in Khuzestan. The major water losses in furrows are due to infiltration, however, measuring amount of infiltration is time consuming and costly. So, it is important to use a method for determining and management of infiltration in the farms. Regarding that, fuzzy logic (FL) and artificial neural network (ANN) were evaluated for five sugarcane industrial farms in Khuzestan. ANN with 12 scenarios (LogSig and TanSig activation functions with 3, 5 and 7 neurons in hidden layer for both N and H farrows) and FL with 8 scenarios (TriMF and GaussMF with 2 and 3 membership functions for both N and H farrows) were studied. Results showed that ANN with LogSig-5 activation function had the best error (RMSE=0.12 m3.m-1), accuracy (NRMSE=0.037) and efficiency (d=0.99 and EF=0.99) for simulation of infiltration in H furrows. ANN with TanSig-5 had the best error (RMSE=0.20 m3.m-1), accuracy (NRMSE=0.11) and efficiency (d=0.99 and EF=0.99) for simulation of infiltration in N furrows. FL had weak accuracy and efficiency for simulation of infiltration in N furrows, however, TriMF-2 had acceptable error (RMSE=1.3 m3.m-1), accuracy (NRMSE=0.052) and efficiency (d=0.99 and EF=0.98) for simulation of infiltration in H furrows. According to comparison of all scenarios, ANN accuracy was about 82% more than FL.